Song Of The South - many projects, much info & discussion thread (Released) — Page 27

OK, 13 whole seconds cleaned up from the cartoon sequence of R3. I wanted to test my deflicker-er, so I did it on a scene with flickering firelight. I first stabilized, then deflickered, then auto followed by manual dust removal, got rid of some of the wear lines, color corrected, then cropped. What does everyone think? Criticisms on workflow, output, etc. appreciated. :)

https://vimeo.com/67764453

Huh. For whatever reason, the frame rate is set to 10, rather than 24...

I think a Lut scan would be best. Also, raw PNGs would be easiest on my end, if that's a format you can output to...

how do you change the bit setting? I think Spaced Ranger is right - it seems we're stuck in 8 bit color.

There's nothing like watching a rock-solid video after having viewed it jittering about! A major improvement (that everyone takes for granted).  :)

After a comparative review (the SOTS-R3-LUT to the SOTS-R3-LUT-PFC-all), I would recommend doing the clean-up in discreet steps, one at a time, with a full render after each step for critical inspection. This will confirm that each process is doing it's job completely, without introducing additional artifacts.

As a first step, to get started on the right track, re-render the uncorrected clip and confirm the rendering is sufficient adjusted reproduce the source identically. If you can't, look for another renderer that can do the job. I bring this up because, when checking out PFClean's "dust removal" (a spacial repair, pictured here), I noticed what looked like macro-blocking in nearby areas. It could be from the rendering ... or it could be from the clean-up ... or from one provoking the other.

In this magnified section, the top is the LUT-capture and the bottom is post-PFClean.

For large debris, like that yellow blob on this single frame, I would overlay a matching replacement area from the previous or next frame (a temporal repair). There are probably many ways to do this same thing in PFClean or maybe they hide all that from you. If you have no control, fixes like this partial clean, with additional smearing, just trades one bit of garbage for another ... needlessly. If so, you should consider using outside programs (there are Avisynth plug-ins that do this sort of this automatically) where PFClean fails to be the perfect solution.

Also, do not necessarily assume that each category of fix must be a one-shot application. Breaking it down to multiple, sub-category fixes might do a better job.

I previously emphasize the word "automatically" because the only "manual" clean-up should be in setting parameters for the program to do the work (otherwise this may end up being your life's work).

Regarding workflow, I've edited and re-ordered what I think would be a good (non-interfering) sequence of fixes:
¤ capture
¤ luminance stabilize
¤ frame motion stabilize
¤ de-spot (single-frame debris)
¤ de-scratch (predictable multi-frame debris)
¤ de-noise
¤ align (non-squared framing)
¤ color correction (film fade)
¤ sharpen (non-haloing)
¤ mask/resize/frame (HD, DVD)

Thanks for the input. I really appreciate your posts here - they are very helpful, at least for me, because I'm learning as I go. Thanks so much for your information and help!

I was hoping no one would notice that yellow blob. I had some problems with it, and took the lazy way out. That was me correcting in a very sloppy, hasty way - I would not correct the final version like that. One of the problems I was running into was the low resolution of the source. To save space, Poita and I had agreed to save the files as a very high-compression video to run tests on, and for me to familiarize myself with PFClean. I imagine (just a guess) that a higher resolution frame will provide an easier go at correcting that error.

Can you describe/illustrate what you mean by macro-blocking? Also, would you elaborate on why you would do luminance stabilization before frame stabilization? Antcufaalb and I have been doing those steps in reverse order...

I'll do another render tomorrow morning, with no settings at all, to see if it is the render in Virtualdub or in PFClean that is causing the artifacts.

ww12345 said:

Thanks for the input. ... because I'm learning as I go.

Glad to help. I'm learning, too (my "practical research").  :)

Macro-blocks are these artifacts (error-generated square boundaries within a picture):

Once you know what to look for, they practically jump out at you. They are generated when a video is rendered at too high compression (or if areas of the picture are excessively smooth). The render software works on squares of areas and flattens those of too-similar pixels (usually from noise reduction) into uniform-color blocks and/or edges.

One way to prevent this is to reduce the compression of the rendered video, which unfortunately increases it's file size. Another way is to trick the render software, to not flattening smooth areas, by adding mild pixel noise back into the video (!). This is interpreted as detail and is not inadvertently flattened. Clever, eh?

In the case of SOTS, use lower compression settings in your renders until macro-blocking no longer appears (also, look for sub-settings specifically for this issue that might be in your render software).

As for the sequence of fixes, I try to arrange them so that a first fix won't hinder or complicate the second fix. You must learn what the fixes do and how they do it to suspect if the order might be improved.  If not sure, experiment -- rearrange the sequence and see if the result is better.

For example, frame-movement stabilization uses tracking to see how "stationary" target pixels move around. Frame luminance fluctuations might mis-lead that tracking and cause positioning errors. On the other hand, luminance smoothing works on the full frame regardless of the internal picture position. Therefore, I would stabilize the frames luminance, first, to allow the best stabilization of their movement.

RE:"Avisynth script to sample a single frame inside each shot change"
AntcuFaalb said:

All you need to do is use the SCSelect() function.

Thanks for the reference. SCSelect() is not native to Avisynth but it is in the plug-in file removedirt.dll as per this Doom9 forum thread:

Real Life TV Scene Change Detection?
http://forum.doom9.org/showthread.php?p=1193539#post1193539

I was looking for something else I remember coming across, a native function -- ScriptClip() or another of Avisynth's Runtime Functions -- with a small script to do the sampling. (No need to load in additional plug-ins, for the novice user.) But whatever works.

A Disney cell from ww12345's gallery http://originaltrilogy.com/forum/topic.cfm/Song-Of-The-South/post/639161/#TopicPost639161 was used as a same-scene color target:

To produce the color correction, I tried for the Olympic gold. But to assume that a cel is an accurate depiction of what was originally on film is a bad assumption. Animation was often made with over-reaching or just plain weird colors to counter particular film stock limitations to produce the desired colors. Unlike my first attempt some pages back, this attempt is on poita's new LUT-enhanced capture (a better & brighter starting point from the previous capture):

                              LUT capture                                                  cel-guided color correction

I did what I could to pull out the green of the fox's shirt, but that seemed to be to big a "spike" in the color curve that affected parts of his apron. Also, adding what green I could affected the blue of rabbit's trousers. All the while, I kept close watch on the rope and other fox & rabbit color. It ended up not much different from that previous adjustment's correction, but still short of the target (even if that target might be an unrealistic one). It would be great if Zip Doodah (hope you're follow the thread) could post a screenshot of this scene to see how it looks on film (with most of the original color intact).

Anyway, here are the numbers I used. Despite my desire for adjustment in R-G-B sections exclusively (in Histogram), I was forced to make additional tweaks using joined R-G-B adjustments, in Histogram and Hue/Saturation/Lightness, to increase the contrast and color saturation, respectively:

             low  gamma   high  midtone
RED        12        1.2   255        +25
GREEN     7         1.0   255        +20
BLUE        0         0.8   255        +20

                       low  gamma   high  midtone
LUMINANCE      0        1.0    255        -10

SATURATION   +10

I'll try this again on the live action part (the coach scene) to see if a good correction made there also produces a good result when applied blind to the cartoon part.

Interesting that you boosted the saturation, as well as correcting the colors. I've never seen Brer Fox's shirt that green, even in promo merchandise and lobby cards/ephemera. 

I don't think that work you do on the coach scene will work because of the fade differences, but you never know. I'd be interested in trying it. 

Now that I look at that cel, does it look off model to anyone else? Just curious...

Well, I'm glad the consensus is that the cel is not a production cel.  :D

ww12345 said:

I don't think that work you do on the coach scene will work because of the fade differences ...

Yeah, I was thinking that, too. Fortunately, there is the 35mm color-target of the girl (cropped to remove distractions):

Here's something quick to check the one setting approach for a master color correction. On the very beginning shot of the cartoon reel is shot of the girl. So, my color correction:

             low  gamma   high  midtone
RED         0        1.5    255           -5
GREEN     0        1.4    255           -5
BLUE        0        1.2    255           -5

                              LUT capture                                                  35mm-guided color correction

Of course, this may not be accurate for this shot, but it's close enough by eye.  This setting is then applied blind to the cartoon:

Not bad. It could be a little hot in the brighter areas (probably same with the girl for that matter). I'd really like to use 35mm sample shots to know for sure.  ;)

I'm debating on how to make the available work on this. Right now there are 5 1920x1080 prores quicktime files- those are the top quality from the transfer. I had thought of making the same offer here as the fundraiser was to help make the transfer (a DVD plus an HD Mpeg-4 quicktime that fits on a seperate dvd)  but I also like the idea of a few folks here cleaning it up further before making it available. Thoughts?

I also like the idea that if there is this much effort put in that it would be much nicer to start with much better material as this transfer has....

True. Also, maybe we can do a Blu Ray of this, which will retain the full quality.

monks19 said:

True. Also, maybe we can do a Blu Ray of this, which will retain the full quality.

I second that.  I'd definitely consider making something sized for a BD25 available...even without a full menu structure.  The more "breathing room" the better.

Zip Doodah said:

I'm debating on how to make the available work on this. Right now there are 5 1920x1080 prores quicktime files- those are the top quality from the transfer. I had thought of making the same offer here as the fundraiser was to help make the transfer (a DVD plus an HD Mpeg-4 quicktime that fits on a seperate dvd)  but I also like the idea of a few folks here cleaning it up further before making it available. Thoughts?

 

I also like the idea that if there is this much effort put in that it would be much nicer to start with much better material as this transfer has....

 

I would think that two separate releases - one raw and one cleaned up, both on BD-25s would be more than sufficient. I don't think people would be opposed to helping pay for the transfer, either, though I'm not sure what the legality of that is. Maybe they could pay for the materials and the time, and not the movie itself?

Also, I'd like to volunteer myself and my time to helping clean up the transfer.

What do you mean about starting with better material? You mean as far as the 16mm version is concerned vs. the 35mm version?

This seems like a good route to go. Perhaps an offer of the 'raw' transfer in SD for now and the cleaned up version when finished as a bdr and a standard disc.